A Deep-Learning Approach to a Volumetric Radio Environment Map Construction for UAV-Assisted Networks

Table 2

Altitude-aware Spider-UNet model parameters for Sliced-VREM construction.

Model parameters	Values
	Encoder	Decoder	Interconnector
Parallel UNet stack	3
Basic 2D-Building blocks	2D Conv.; 2D Conv.; 2D Max-pool (2)	Transposed 2D Conv.; Concatenate; 2D Conv.; 2D Conv.; Stride (2)	Conv-LSTM (bi-directional)
Layers	4	4	1
Filters	(64, 64) ⟶; (128, 128) ⟶; (196, 196) ⟶; (256, 256)	(256, 256, 256); (196, 196, 196); (128, 128, 128) (64, 64, 64)	512
Input resolution for single stack	(128, 128, 1) + (128, 128, 1)	—	—
Output resolution for single stack	(128, 128, 1) + (128, 128, 1)	—	—
Kernel size	4
Activation	LeakyReLU (alpha = 0.2)

as an output 2D convolution layer with a single channel.